Research of Simulated Annealing for BP Network and Its Application

In order to avoid overfitting and fall into local minima, a hybrid algorithm combining simulated annealing algorithm and BP algorithm is used to train neural network. The strong global optimization ability of simulated annealing algorithm and the fast local optimization ability of BP algorithm are combined to ensure that the global minimum point can be found more accurately without reducing the learning rate of the network. The model is applied to Taobao e-commerce sales forecast and the results are satisfactory.